Zone Scout
Multi-agent lead intelligence system with an automated enrichment layer over Google Maps Places API — verified contacts, reviews, business metadata, and social profiles per validated lead.
Pipeline
The problem
Maps list businesses but do not judge them. Building a qualified local lead list means opening every listing by hand, checking whether it matches your criteria, and hunting for contacts and social profiles separately.
Architecture
End-to-end flow, stage by stage.
- 01
Map screenshot/ZIP
- 02
Gemini geocoding
- 03
Llama audit
- 04
Places enrichment
- 05
Streamlit UI
How it works
Zone defined by ZIP or screenshot
A search boundary can be typed as a ZIP code or dropped in as a map screenshot, which Gemini 1.5 Flash reads visually before Google Maps geocoding converts it into a precise geographic zone.
Reasoning agent audits every lead
Llama 3.3 70B on Groq reviews each discovered business against the user's stated criteria and rejects the ones that do not qualify, so filtering is a reasoning step rather than a keyword match.
Automatic enrichment
Surviving leads gain Places API details, reviews, hours and verified contacts, plus context-aware social links — LinkedIn for tech firms, Instagram for retail — collapsing a multi-hour research task into minutes.
Streamlit delivery
The whole flow ships as a Streamlit app with dotenv-managed keys, deployed publicly so it can be tried without setup.
- Accepted either a ZIP code or map screenshot as the search boundary, using Gemini 1.5 Flash vision plus Google Maps geocoding to translate visual intent into a precise geographic zone.
- A Llama 3.3 70B reasoning agent on Groq audited discovered businesses against the user's criteria before they entered the result set.
- Enriched validated leads with Places API details, reviews, hours, contacts, and context-aware social links, turning a multi-hour research task into a minutes-long workflow.
By the numbers
Lead quality lift
Enrichment sources per lead
higher is betterLeads audited automatically (%)
higher is betterTime per zone (minutes)
lower is betterWhat it ships with
- Multi-agent vision plus reasoning pipeline
- ZIP code or map screenshot input
- Criteria-based lead auditing
- Verified contacts, reviews and hours
- Category-aware social profile generation
- 1-2→4+
- Enrichment sources per lead
- Manual→Zero-touch
- Lead validation pipeline
Stack
More projects
Computer Vision
Multi-Modal Vision-Language Model for Cancer Diagnosis
Cross-modal pipeline correlating high-dimensional medical imaging with unstructured biomedical literature, with a Grad-CAM + Gradio explainability interface for clinical interpretability.
Agentic AI
Pitch Deck Debater
Simulated investor boardrooms with VC, CTO, and Product Lead personas, an adversarial Chain-of-Thought feedback loop, and consensus-voting logic that resolves inter-agent hallucinations.